Scholarlink

<p>Plastic waste management remains a critical environmental challenge, particularly in developing countries where low-cost and scalable sorting technologies are required. This study presents the development and evaluation of an automated plastic bottle separation system integrating a convolutional neural network (CNN) with a mechanically simple sweep-arm mechanism. The system is designed to classify and separate high-density polyethylene (HDPE) and polyethylene terephthalate (PET) bottles conveyed on a belt using RGB imaging and a MobileNetV2-based object detection model. A custom dataset of HDPE and PET bottles was developed and used to train the CNN, achieving mean Average Precision (mAP) values exceeding 80% for both HDPE and PET classes at standard Intersection over Union (IoU) thresholds. The trained model was deployed in a prototype consisting of a conveyor belt and a single-actuator sweep arm, enabling material-based diversion at the end of the conveyor. Beyond algorithmic evaluation, this work presents an analytical and experimentally validated model for determining the minimum inter-bottle spacing required for reliable separation, which is critical for ensuring stable operation and preventing mis-separation. The obtained waste-handling capacity of the system was analytically estimated to range from 32.4 to 103.6&#xa0;kg/h, depending on the bottle material composition. The results demonstrate that a mechanically simple, CNN-assisted separation system can achieve reliable plastic bottle sorting with competitive throughput, offering a cost-effective alternative to more complex robotic or spectroscopic sorting systems. This work highlights the importance of integrating mechanical design considerations with machine vision performance for practical waste separation applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance evaluation of a plastic bottle separation system using MobileNetV2 CNN and a simple sweep arm mechanism

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • M. Qibthiyah,
  • R. Dhelika

摘要

Plastic waste management remains a critical environmental challenge, particularly in developing countries where low-cost and scalable sorting technologies are required. This study presents the development and evaluation of an automated plastic bottle separation system integrating a convolutional neural network (CNN) with a mechanically simple sweep-arm mechanism. The system is designed to classify and separate high-density polyethylene (HDPE) and polyethylene terephthalate (PET) bottles conveyed on a belt using RGB imaging and a MobileNetV2-based object detection model. A custom dataset of HDPE and PET bottles was developed and used to train the CNN, achieving mean Average Precision (mAP) values exceeding 80% for both HDPE and PET classes at standard Intersection over Union (IoU) thresholds. The trained model was deployed in a prototype consisting of a conveyor belt and a single-actuator sweep arm, enabling material-based diversion at the end of the conveyor. Beyond algorithmic evaluation, this work presents an analytical and experimentally validated model for determining the minimum inter-bottle spacing required for reliable separation, which is critical for ensuring stable operation and preventing mis-separation. The obtained waste-handling capacity of the system was analytically estimated to range from 32.4 to 103.6 kg/h, depending on the bottle material composition. The results demonstrate that a mechanically simple, CNN-assisted separation system can achieve reliable plastic bottle sorting with competitive throughput, offering a cost-effective alternative to more complex robotic or spectroscopic sorting systems. This work highlights the importance of integrating mechanical design considerations with machine vision performance for practical waste separation applications.